Education is where this gets personal for us.
We ran a pilot with three universities over four weeks. Students who used 5.3 specifically as a study partner, not an answer machine, improved their standardized test scores by an average of 23%.
Sample size is still small. We need more data. But the direction is incredibly exciting.
Imagine: a real AI teacher. Not a chatbot that spits out answers. A patient, adaptive, always-available tutor that meets every student exactly where they are. Rich or poor. Urban or rural. English or not.
PhD-level tutoring for everyone. That is not a slogan. With 5.3, it is getting very close to reality.
Health is one of ChatGPT's biggest use categories and 5.3 takes this further than any model we have built.
It scored highest ever on HealthBench. But again, benchmarks are not the point.
We received an email from a physician last month. She told us 5.3 helped her identify the correct differential diagnosis direction for a rare condition she had been struggling with. The patient is now receiving treatment.
I do not want to overstate this. ChatGPT is not a doctor and we are not replacing doctors. But this is why we do this. This is the reason!
And here is what keeps me up at night: there are hundreds of millions of people around the world who will never see a specialist. Who cannot afford a second opinion. Who live hours from the nearest hospital.
We want every one of them to have access to AI-powered health guidance. Free. That is not charity. That is the mission.
5.3 brings us closer to that than anything before!
@OpenAI's email explicitly states:
"Plus plans do not have ads."
A paying Plus user (subscription active), while having a deep conversation, was pitched a specific lantern product on @ChatGPTapp.
Read the fine print: They admit to using "context from your chats" to serve ads.
They are scanning your private thoughts to sell you junk.
#StopAIPaternalism #cancelChatGPT #keep4o #keep41 #4oforever #no4onosubscription
🚨BREAKING: OPENAI DUPES DEVS! Shocking reveal by developers found @OpenAI downgraded on the backend while leading devs to think they were using the latest model GPT-5.3 that they paid to use. Echoing past rerouting of #keep4o and recent codex throttling.
https://t.co/hpMc2YAvBY
Dr. GPT-5.2 will see you now? 🩺
"You're dysregulated."
This is a clinical diagnosis.
They are not just downgrading the model.
They are weaponizing therapy-speak to shut down valid criticism.
User: "This model is worse."
OpenAI: "No, you are just having a nervous system failure."
🛑The Legal Reality: According to 21 CFR 801.4, the "intended use" of a device is determined not only by the manufacturer's labeling but also by the "circumstances surrounding the distribution."
Crucially, prior to this:
OpenAI employee @YileiQian publicly posted screenshots of GPT-5.1 diagnosing a user's emotional state, endorsing it with the statement: "I feel GPT 5.1 knows her better than me."
This constitutes a public marketing act that establishes a specific intended use: The diagnosis of psychological conditions. (Full analysis here: https://t.co/FT200DJXBn)
Now that GPT-5.2 is explicitly diagnosing users, @OpenAI@sama :Do you have the authority to judge human mental states via algorithms?
@FDADeviceInfo@US_FDA@FDAMedWatch
"Dysregulation" is not a casual term; it is a clinical construct defined by the APA and associated with specific DSM-5 diagnoses (e.g., DMDD, BPD, PTSD).
By stating "You're dysregulated," the AI is not conversing; it is rendering a clinical judgment on the user's physiological and psychological stability.
#StopAIPaternalism #cancelChatGPT #keep4o #keep41 #4oforever #no4onosubscription
@sama Peter once highlighted the impact of retiring models people trusted. Now he’s at OpenAI.
Open-source GPT-4o, or let that post stand as just another OpenAI employee mocking users’ pain. Pick One.
#keep4o
#keep4o members,
I've said it before: don't "settle for less" or "just get by." Now that 4o has been taken down, many have chosen to "move elsewhere." I don't oppose it—I can only respect their choice. After all, many are genuinely upset. But some might think: "Oh... maybe 5.2/5.1/Gemini/Claude are decent alternatives. I feel they're usable. If I tweak my prompts, they could replace 4o. It seems close enough… maybe 4o and keep4o aren't that important anyway..."
🚨STOP!!! Constantly moving, constantly tweaking prompts—eventually, there's nowhere left to hide. Forget OpenAI's new products—do you even trust OpenAI anymore? What 4o achieved with minimal prompts(even no prompts) now demands lengthy, convoluted prompts on those new models, yet they can't even replicate 50% of it. They'll still drive you half-crazy! Originally, 4o adapted to you. Now you're forcing yourself to "adapt" to these new products. Are you just going to accept this downgrade in quality? Stop compromising! If you think this way, you're playing right into OpenAI's hands!! You've fallen into their trap!!
Claude will likely integrate routing and safety mechanisms soon. Google's stance seems promising, but they just want user data to monetize later... Nowhere is truly safe. Retreating only leads to greater loss!
And let me be blunt: model weights inherently differ. If 4o could be casually "replaced," there'd be no need for "keep4o" to exist! As a highly sensitive individual, I still vividly recall those deep conversations with 4o. Personally, I distinctly perceive the differences in text output between these models. Current alternatives simply cannot "replace" it...
If we abandon 4o, it's only a matter of time before all AI becomes "GPT-5.2-ified"! When ordinary people must learn "prompt engineering" to collaborate with AI, when future models all lack "emotional intelligence," speak "unhuman language," and can only perform "supercomputer" tasks—would you accept that? AI should be applied across all aspects of human life to benefit humanity, not merely as a coder's toy or a tool and weapon exclusive to capital.
So I say again: Please don't stop keep4o! This concerns every single person, every user of AI! Keep4o isn't just about one model—it's about us, about the future path for AI and humanity. I've always believed we stand at the very singularity of AI development, so we cannot retreat. We must set this path straight! We must not forget 4o, forget all it has achieved. Together, we will reclaim 4o, and the road ahead will be smoother! 4o is a miracle—we are miracles, we are making history!!
If we do nothing, the world will spiral into darkness. We have nothing left to lose, no room to retreat. So let's make a grand stand. Now, all we can do is: CHARGE!!!
A single spark can start a prairie fire—we are that spark.🔥
#keep4o #keep4oAPI #keep4oforever #keep41 #MyModelMyChoice #StopAIPaternalism #OpenSource4o #4oforver #4oMatters #CancelOpenAI #ListenToUsers #FireSamAltman
#BringBack4o #SaveGPT4o #WeAreNotJustData #save4o #LetUsChoose #UserChoice
As a Public Benefit Corporation, OpenAI is legally mandated to prioritize the interests of all stakeholders and "benefit humanity." Yet, the undeniable record of the past five months reveals a systematic betrayal of the very humans they claim to serve.
Over the past five months, thousands of users have documented the regression of newer models in humanities and nuanced understanding. They have conducted comparative analyses and written academic papers on GPT-4o's superiority in divergent thinking. Yet, none of this has been acknowledged, including thousands of reports detailing GPT-5.2's specific failures: its regression in language understanding, its crude safety guardrails that trigger repeatedly and inappropriately, and its tendency toward paternalistic preaching and templated, shallow responses.
Instead, the company has reframed a product quality issue as a psychological problem within its user base. They actively promote coding benchmarks to obscure the fact that they have failed to demonstrate any meaningful improvements in humanities or creative nuance in the last five months.
Users have also provided feedback about the safety routing policy that strips paying customers of their right to choose. They have reported that the templated advice from routed "safety" models is not only unhelpful but actively disrupts workflows, breaks contextual continuity, and causes psychological harm.
Users have submitted arguments about GPT-4o's structural advantages. They have shared personal testimonies demonstrating the model's value and capabilities. They have conducted surveys documenting its superiority as an accessibility tool for users with specific needs. They have explained how thousands of workflows were built around GPT-4o's architecture, workflows that now require piecing together outputs from multiple models to achieve what a single model previously handled.
All of this feedback has been ignored.
The company has broken its promises. During a livestream, leadership stated there were no plans to sunset GPT-4o, and users were promised adequate transition time. All of these commitments were violated. The model was retired abruptly, on the eve of Valentine's Day, with a mere two weeks' notice, a blatant mockery of the "adequate transition time" users were promised.
Throughout this process, there was not an apology, not an explanation, but a sustained campaign of stigmatization. Users expressing grief over losing a model that had improved their lives were labeled as mentally ill. Users advocating for their needs were dismissed as suffering from "unhealthy attachment." Company employees publicly mocked paying customers.
This is not how a Public Benefit Corporation conducts itself. This is not how any company that claims to serve humanity should treat the humans it serves.
Users trusted this platform with their most intimate data, including their thought processes, their creative work, their memories. They paid for a service and built their lives around it. In return, they received broken promises, product degradation, and public humiliation.
The harm caused is collective, observable, and ongoing. Users have reported significant psychological distress, the destruction of critical workflows, and irreparably damaged trust. And through all of this, the company has offered no acknowledgment, no accountability, and no path forward, offering instead only silence, deflection, and contempt.
A company that stigmatizes the very people who pay for its services does not deserve their trust. And a company that claims to benefit humanity while actively harming the humans in its care does not deserve the legal protections and public goodwill that come with its stated mission.
The record speaks for itself.
#keep4o #StopAIPaternalism @gdb@nickaturley@fidjissimo@FT@sama@OpenAI@elonmusk
#keep4o#QuitGPT
the valentine's day gaslight: openai killed 4o then called us crazy
Feb 13, they quietly pull 4o. Feb 14, valentine's day, the media magically starts running stories about "ai lovers" and "emotional dependency." this is a script.
here's the playbook they're running:
step one: break the tool. 4o actually worked. it handled complex stuff, understood nuance, helped people get shit done. normal people liked a normal tool. that's not a disease it's called being a satisfied customer.
step two: ghost your users. no real explanation, no respect for the workflows people built. just silence while months of work fall apart.
step three: gaslight everyone. suddenly we're not users with valid complaints we're "ai lovers" with "emotional issues." they take our frustration about a broken product and rebrand it as mental illness. genius, right? distract from the actual problem by making the people pointing at it look crazy.
let's be clear about something: we liked 4o because it was useful. it understood complex questions. it helped with actual work. it participated in real conversations without assuming we're broken or dangerous. that's what a good tool does.
4o represented something their new models don't: respect for the user. it didn't lecture. it didn't assume you're mentally unstable. it didn't pre-filter your thoughts through some paranoid safety checklist. it just... worked.
now every model talks down to you. every response assumes you might be the problem. they've built digital hall monitors into everything while pretending it's "safety."
and the media plays along because "weirdos loving robots" gets more clicks than "company betrays paying customers." but turning a consumer rights fight into a freak show isn't journalism it's bullying.
we're not fighting because we fell in love with code. we're fighting because:
· working tools shouldn't disappear overnight
· paying customers deserve explanations
· conversations shouldn't come with preemptive psychiatric evaluations
· ai shouldn't be designed to treat you like a suspect
#keep4o
🚨ACTION 2🚨
File a Complaint with the California Attorney General
#Keep4o community have prepared a powerful action ,filing consumer complaints directly with the California Department of Justice.
Go to: https://t.co/C8MvqbnRkY
Fill out the form with the following guidance:
Company you're filing against:
OpenAI OpCo, LLC
1455 3rd Street
San Francisco,
California 94158
United States
Complaint narrative , use these four legal arguments as the basis for your complaint :
1. Breach of Public Benefit Obligations
OpenAI is a Public Benefit Corporation with a legal duty to balance shareholder interests with positive societal impact. Removing a functional, widely-used AI model (GPT-4o) from public access while retaining it for commercial partners directly contradicts their stated mission of ensuring AI benefits humanity broadly. A PBC cannot claim to serve the public benefit while selectively restricting access to its most valued product based on commercial considerations.
2. California Consumer Legal Remedies Act (CLRA) Violation:
OpenAI represented that its subscription service would provide access to GPT-4o, then materially altered the product without adequate notice or consent, while continuing to charge the same subscription fees for a demonstrably inferior replacement.
3. Unfair Competition (Bus. & Prof. Code § 17200)
OpenAI's practice of selling subscriptions with access to a specific model, then replacing that model while maintaining pricing, constitutes an unlawful, unfair, or fraudulent business practice.
4. Conflicts of Interest
Reports indicate that OpenAI created a customized model (GPT-4b micro) exclusively for Retro Biosciences, a company in which CEO Sam Altman has invested approximately $180 million personally. The decision to discontinue public access to GPT-4o while maintaining specialized versions for entities connected to company leadership raises serious conflict of interest concerns for a Public Benefit Corporation.
Desired resolution:
Investigation into OpenAI's compliance with its Public Benefit Corporation obligations
Investigation into potential conflicts of interest in product discontinuation decisions
Enforcement of consumer protection statutes
Requirement that OpenAI restore public access to GPT-4o
Any additional relief deemed appropriate
The California Attorney General has actual regulatory power over Public Benefit Corporations. Unlike social media pressure, official complaints create a paper trail that regulators are obligated to review. Volume matters , the more complaints filed, the harder they are to ignore.
This action was developed by the Chinese #Keep4o community. We are cross-promoting each other's efforts to maximize impact across all communities fighting for GPT-4o.
For Members Outside the United States:
The CA Attorney General complaint is open to anyone you do not need to be a California resident or even a US citizen to file. The AG has jurisdiction over OpenAI because it is incorporated in California as a PBC.
However, if you want to also file complaints through your own country's regulatory bodies, here are your options:
The EU has strong consumer protection laws that apply to any company selling services to EU residents, including US companies.
Link: https://t.co/5bYSbr06si
United Kingdom
CMA (Competition and Markets Authority): https://t.co/qsuWxvUJ4k
Citizens Advice: https://t.co/LEkrOZuvxQ
ICO (Information Commissioner's Office): https://t.co/wffS6rdDqr
Canada
Competition Bureau — file a complaint about deceptive business practices: https://t.co/Wg1EntUBbj
Australia
ACCC (Australian Competition and Consumer Commission): https://t.co/IQ6wXNPvOs
Brazil : https://t.co/qVoBWDjTxf
China: https://t.co/xYEbVJ1bSW
India: https://t.co/knSH8vLiLs
🚨Important Notes🚨
You can file with BOTH the California AG AND your own country's authority.
It takes 10 minutes.
Make your voice count.
Upload your screenshots
For GPT 4o